Who liquidity providers lose to

The fee is not the return

Every swap pays the provider a fee. Some come from traders who know the price is about to move. The vault buys the other side, the price moves, and the position is worth less one hour later. That loss is the markout. The useful number is the fee earned per dollar of markout loss, because a vault can quote a high fee APR and still lose money to informed flow.

The split comes from the NVDA, SPCX and TSLA pools. B2 is a plus or minus 2% band, B5 is plus or minus 5%, and each cell is the fee earned per dollar lost over a one hour markout. A ratio below 1 means it pays informed traders more than it collects in fees.

Two patterns survive the noise. NYSE regular hours are the most toxic session for every pool: SPCX collects $0.78 per dollar lost and TSLA collects $0.19 at the B2 band. Weekends are the least toxic: SPCX reaches $7.07 and TSLA reaches $1.69. The second pattern is that the pool matters as much as the session. SPCX swings from 0.78 in NYSE hours to 7.07 on weekends at the same band, while TSLA never clears 1 in regular hours or weeknights at either band.

Session timing does not beat a static band

A ratio that looks good by session is not enough to build a schedule. The replay tested the obvious version: run a tight band during NYSE hours and a wide band outside them. It used 5.1 million swaps from 2026-08-24 to 2026-10-17 on NVDA, SPCX and TSLA, with $1,058M, $697M and $35.8M of volume. Each strategy ran at $100,000 and at $1,000,000, with a fee-only cost model and a pool-impact cost model.

The hours-aware strategy won 0 of 3 tokens at both sizes. It lost to the best static band on every token. The result is not a small miss: on TSLA, strategy C returns -13.9% against holding while a static B5 band returns 42.0% at the same size. The wide static bands mostly won because the vault held half its capital in USDG while the tokens fell, so the comparison is against a falling asset, not proof of a fee edge.

The control is the part that closes the case. The replay shuffled the session labels, assigning the real schedule to random times, and the shuffled schedule beat the real one in 64 of 64 cells. A pattern that wins after its labels are destroyed is not a timing signal. An inverse schedule did win 12 of 12 cells after the fact, then faded to 8.4% on NVDA and -11.7% on SPCX. That is a warning about post-hoc selection, not a strategy.

What the data did support

The fee to markout ratio separated pools and bands even when the session clock did not. The replay chose a different best static band for each token: B5 on NVDA and TSLA, B2 on SPCX. That is why Uptick does not sell a weekend vault or an hours-aware vault. It measures the ratio per pool, chooses the band per token, and publishes fee APR beside kept APR.

The 5.1 million swaps set the size of the effect. SPCX at B2 pays $0.78 in NYSE hours and $7.07 on weekends; TSLA pays $0.19 and $1.69. A schedule that moved capital between those sessions would have to beat the loss it takes in the bad one. The test says it does not.

What this means for a holder

A vault page shows fee APR next to the APR the vault actually kept after markout and rebalancing cost. The deposit gate reads the kept figure, so a pool with a high fee APR and a weak kept APR stays closed. The session split is published because it explains why the design uses a wide band and a fixed recentre time instead of a market-timing rule.

When a vault quotes a return, the number is the result of a replay on that pool at that vault's own cap. It is not an annualised fee rate, and it is not a claim that weekends always pay. The 54 day window contains the evidence, and the holder can check the same fee-to-markout figures before depositing.